Bug #53458 Infinite out-of-order buckets after master failure in drop6

• Submitted: 6 May 2010 8:26
• Modified: 24 May 2010 13:08
• Reporter: Jonas Oreland
• Status: Closed
• Category: MySQL Cluster: Cluster (NDB) storage engine
• Severity: S3 (Non-critical)
• Version: drop6
• OS: Any
• Assigned to: Jonas Oreland
• CPU Architecture: Any

[6 May 2010 8:26] Jonas Oreland

Description:
In drop6, there is a race-condition so that
in case of master failure, a SUB_GCP_COMPLETE_REP
might in some circumstances be sent twice from same node.

This makes ndbapi be very confused.

---

Note: This problem is fixed in >= telco-6.2 during introduction of
micro-gcp

How to repeat:
send COPY_GCIREQ to not all nodes and then kill master.
new master will then send COPY_GCIREQ to all nodes,
and the ones that get it twice, will also emit SUB_GCP_COMPLETE_REP
twice.

Suggested fix:
Check for condition, and don't emit twice.

[24 May 2010 13:08] Jonas Oreland

Fixed in drop6,
no changelog needed